[time-nuts] Gate propagation delay jitter

Magnus Danielson cfmd at bredband.net
Tue Apr 10 17:50:57 EDT 2007

From: Dr Bruce Griffiths <bruce.griffiths at xtra.co.nz>
Subject: Re: [time-nuts] Gate propagation delay jitter
Date: Wed, 11 Apr 2007 09:32:26 +1200
Message-ID: <461C026A.3080402 at xtra.co.nz>

> > One of the wizards on the fpga newsgroup has a war story about clock jitter 
> > due to outputs near the clock input switching.  I forget the details.  It may 
> > have been a single ended clock input.
> >   
> Hal
> Yes ground bounce can play havoc with the effective switching thresholds.
> One would expect this effect to be much worse with single ended clocks.

You would also keep clear of the MGT/GTP pins unless you are driving/receiving
very slow signals rarely. Ground-bounce is a real issue with these things and
as you are into large pin-counts, it becomes a real mess. You do want to plan
your bypass capacitance with care (both in-board planes and on-board chip-
caps). They have also started to do things that I and probably others proposed,
put bypass caps on the BGA carrier PCB. That alone greatly simplifies the

Single ended clocks is indeed more sensitive, but many people forget that
balanced traces are more sensitive when you break the balance, such as in a
non-symmetric bend or when you have a non-symmetric (often unintentional)
coupling to a nearby trace. In one design we had a bent balanced ECL clock, but
at excessive stress-testing (running 0x0000/0xffff patterns all the time) we
had sever clock troubles. I troubleshooted it by doing a TDT, using the
reciprocity and sensing the coupling to suspected lines. It was the bending
where the fault was. For the next layout time we had clearence around that
clockpaths bendings and the problem have not been seen since! A balanced ECL
signal is still a small signal compared to the full swing of 3.3V CMOS and with
the high current rate thus drivers where set for. Balancing alone isn't a
magical bullet, it is a tool among others but still needs to be handled with

Also, that's why you have a TDR/TDT capable scope collecting dust in the lab.
When you need it, you bloody well need THAT scope. It has payed its price many
times over.


More information about the time-nuts mailing list